Target Trial Emulation in Ageing Research

Specify the trial before analysing the records

Researchers first describe a hypothetical trial, then map its protocol onto available observational data. The protocol specifies eligibility, treatment strategies, assignment, outcomes, follow-up, the causal effect of interest and the analysis. Differences between the ideal protocol and its implementation should be explicit. This framework concerns study design; a large database or sophisticated statistical model alone does not establish causality. See the foundational framework.

Align eligibility, treatment assignment and time zero

Eligibility assessment, assignment to a strategy and the start of follow-up must align. Defining treatment using future survival can create immortal time bias: participants must remain alive long enough to qualify as treated. The resulting apparent survival advantage may be a design artefact. This problem and related timing errors are examined in the methodological paper on target trials and immortal time.

For a hypothetical ageing study, suppose researchers label people as treated only after six months of prescriptions, but count survival from the first prescription. Those qualifying as treated necessarily survived the qualifying period. A protocol comparing initiation at a common eligible starting point avoids that particular future-survival requirement. It still needs adjustment for differences between groups. If treatment may begin within a grace period, simply classifying people by eventual treatment is insufficient; the design and analysis must account for that interval.

See Immortal Time Bias in Longevity Studies.

What emulation cannot guarantee

Causal interpretation still requires adequate control of confounding, meaningful treatment definitions and sufficient overlap between comparison groups. Missing or inaccurate records can undermine these conditions. Emulation does not recover unmeasured information or guarantee that observational groups are comparable. The foundational paper discusses these limits.

Reading an ageing-research application

The following are appraisal questions, rather than findings from a particular intervention study:

• Does the dataset measure frailty, functional status, disease severity and reasons for treatment selection adequately?
• Are people with multimorbidity represented, and to whom could the result reasonably apply?
• Does the outcome describe mortality, disability, a specific disease or a biomarker change?
• For a nonfatal outcome, how does the analysis handle death before that outcome?
• Are absolute effects, uncertainty intervals and sensitivity analyses reported?

Human observational evidence should remain labelled as such. Animal survival experiments and laboratory studies provide different evidence and cannot substitute for a human treatment comparison. A hypothetical protocol supplies no intervention efficacy result. Likewise, an estimated effect on one disease or biomarker should not be described as proof of slower ageing or longer human lifespan.

For interpretation, see Competing Risks in Ageing Research and Surrogate Endpoints vs Clinical Outcomes.

References

1. Hernán, M. A., et al. (2016). Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses . Journal of Clinical Epidemiology . https://pmc.ncbi.nlm.nih.gov/articles/PMC5124536/
2. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available
3. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available
4. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available
5. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available